Abstract
BACKGROUND: Monitoring and evaluating dynamic changes in brain states during electroencephalography (EEG) neurofeedback training (NFT) for post-traumatic stress disorder (PTSD) patients remains challenging when using traditional methods. METHOD: This study proposes a novel Process Noise Dynamic Adaptation-Mesoscale Mesonetwork Network (PNDA-MMNet) model, which improves upon conventional techniques by establishing a discrete linear dynamic model of the NFT process. The model utilizes a mesoscale intermediate network architecture to create a brain state observation matrix, computes the brain state transition matrix, and applies fuzzy rules for dynamic adaptive noise processing. This maximizes the separability between brain state transitions during NFT and resting states. RESULTS: The proposed model achieves a brain state identification accuracy of 0.7428 ± 0.12 (area under the curve, AUC = 0.84), significantly outperforming conventional algorithms. Interpretations of the model indicate that continuous NFT reduces functional connectivity within the motor cortex, with stronger suppression in the right hemisphere compared to the left. Additionally, it reveals decreased activity in the occipital cortex, particularly in the left occipital region, where inhibition increases radially from the midline. Notably, the connectivity between the motor and occipital cortices remains stable throughout the training process. These connectivity changes reflect NFT-induced modulation of cortical activity and are consistent with known neurophysiological patterns in PTSD, highlighting their potential relevance to therapeutic mechanisms. CONCLUSION: This research introduces a more effective approach for real-time monitoring and evaluation of PTSD patients' brain states during NFT, offering a quantitative method for assessing treatment efficacy and guiding therapeutic interventions.